Generating the American Shoulder and Elbow Surgeons Score Using Multivariable Predictive Models and Computer Adaptive Testing to Reduce Survey Burden

Abstract

Background: The preferred patient-reported outcome measure for the assessment of shoulder conditions continues to evolve. Previous studies correlating the Patient-Reported Outcomes Measurement Information System (PROMIS) computer adaptive tests (CATs) to the American Shoulder and Elbow Surgeons (ASES) score have focused on a singular domain (pain or physical function) but have not evaluated the combined domains of pain and physical function that compose the ASES score. Additionally, previous studies have not provided a multivariable prediction tool to convert PROMIS scores to more familiar legacy scores. Purpose: To establish a valid predictive model of ASES scores using a nonlinear combination of PROMIS domains for physical function and pain. Study Design: Cohort study (Diagnosis); Level of evidence, 3. Methods: The Military Orthopaedics Tracking Injuries and Outcomes Network (MOTION) database is a prospectively collected repository of patient-reported outcomes and intraoperative variables. Patients in MOTION research who underwent shoulder surgery and completed the ASES, PROMIS Physical Function, and PROMIS Pain Interference at varying time points were included in the present analysis. Nonlinear multivariable predictive models were created to establish an ASES index score and then validated using “leave 1 out” techniques and minimal clinically important difference /substantial clinical benefit (MCID/SCB) analysis. Results: A total of 909 patients completed the ASES, PROMIS Physical Function, and PROMIS Pain Interference at presurgery, 6 weeks, 6 months, and 1 year after surgery, providing 1502 complete observations. The PROMIS CAT predictive model was strongly validated to predict the ASES (Pearson coefficient = 0.76-0.78; R2 = 0.57-0.62; root mean square error = 13.3-14.1). The MCID/SCB for the ASES was 21.7, and the best ASES index MCID/SCB was 19.4, suggesting that the derived ASES index is effective and can reliably re-create ASES scores. Conclusion: The PROMIS CAT predictive models are able to approximate the ASES score within 13 to 14 points, which is 7 points more accurate than the ASES MCID/SCB derived from the sample. Our ASES index algorithm, which is freely available online (https://osf.io/ctmnd/), has a lower MCID/SCB than the ASES itself. This algorithm can be used to decrease patient survey burden by 11 questions and provide a reliable ASES analog to clinicians.